val_loader, _ = loading_data(target_data_path, mode='val')
net = CSRNet().to(device)
optimizer = optim.SGD(filter(lambda p: p.requires_grad, net.parameters()),
lr=lr,
momentum=0.9)
criterion_dens = nn.MSELoss(size_average=False)
criterion_count = nn.L1Loss(size_average=False)
# 开始训练和验证
best_mae = sys.maxsize
for epoch in range(start_epoch, epoch_num + 1):
print('Epoch {}/{}'.format(epoch, epoch_num))
# 训练阶段
optimizer = update_lr(optimizer, epoch, steps, decay_rate)
net.train()
running_loss = 0.0
running_mse = 0.0
running_mae = 0.0
totalnum = 0
for idx, (image, densityMap) in enumerate(train_loader):
image = image.to(device)
densityMap = densityMap.to(device)
optimizer.zero_grad()
duration = time.time()
predDensityMap = net(image)
predDensityMap = torch.squeeze(predDensityMap)
densityMap = torch.squeeze(densityMap)
loss = criterion_dens(predDensityMap, densityMap)
if __name__=="__main__":
# argument parsing.
args = parser.parse_args()
cfg = Config(args) # configuration
model = CSRNet().to(cfg.device) # model
criterion = nn.MSELoss(size_average=False) # objective
optimizer = torch.optim.Adam(model.parameters(),lr=cfg.lr) # optimizer
train_dataloader = create_train_dataloader(cfg.train_dataset_root, use_flip=True, batch_size=cfg.batch_size)
test_dataloader = create_test_dataloader(cfg.test_dataset_root) # dataloader
min_mae = sys.maxsize
min_mae_epoch = -1
for epoch in range(1, cfg.epochs): # start training
model.train()
epoch_loss = 0.0
for i, data in enumerate(tqdm(train_dataloader)):
image = data['image'].to(cfg.device)
gt_densitymap = data['densitymap'].to(cfg.device) * 16# todo 1/4 rescale effect때문에
et_densitymap = model(image) # forward propagation
loss = criterion(et_densitymap,gt_densitymap) # calculate loss
epoch_loss += loss.item()
optimizer.zero_grad()
loss.backward() # back propagation
optimizer.step() # update network parameters
cfg.writer.add_scalar('Train_Loss', epoch_loss/len(train_dataloader), epoch)
model.eval()
with torch.no_grad():
epoch_mae = 0.0
for i, data in enumerate(tqdm(test_dataloader)):
targetloader_iter = enumerate(target_loader)
best_mae = sys.maxsize
loss_dens_value = 0.0
loss_adv_value = 0.0
loss_D_value = 0.0
running_mse = 0.0
running_mae = 0.0
totalnum = 0
iter_count = 0
for epoch in range(start_epoch, epoch_num + 1):
iter_count = iter_count + 1
print('Iteration {}/{}'.format(epoch, epoch_num))
optimizer = update_lr(optimizer, epoch, steps, decay_rate)
net.train(True)
net_D.train(True)
optimizer_D = adjust_learning_rate_D(optimizer_D, epoch, lr_D,
epoch_num, power)
optimizer.zero_grad()
optimizer_D.zero_grad()
# train G
for param in net_D.parameters():
param.requires_grad = False
# train with source
_, (image, Dmap) = next(trainloader_iter)
image = image.to(device)
Dmap = Dmap.to(device)
